<![CDATA[Background: Digital transformation in diabetes care is accelerating the use of electronic patient-reported outcomes (ePRO) and large language models (LLMs). Evidence remains limited on whether ChatGPT-4 can feasibly analyze neuropathic pain and sleep problems from ePRO data in adults with diabetes. Purpose: This study evaluated the feasibility and quality of ChatGPT-4 when analyzing neuropathic pain and sleep-related ePROs in adults with diabetes. Methods: We conducted a single-centre feasibility study in routine practice using daily and weekly ePROs on neuropathic pain and sleep from adults with diabetes. ChatGPT-4, prompted with distinct clinical personas, generated narrative summaries. A multiprofessional panel rated accuracy, completeness, safety, empathy, and clinical usefulness, and we calculated descriptive statistics and inter-rater reliability coefficients. Results: Across 180 evaluated responses, 71% met the predefined adequacy threshold, with a mean global quality score of 7.1±1.4/10. Safety, accuracy, and clinical usefulness showed the highest ratings, whereas completeness was more variable. Inter-rater reliability for global scores was high (ICC=0.82). Persona framing significantly influenced domain-specific ratings (p<0.05) and higher scores for the physician than nurse personas on several clinically critical items. Conclusion: ChatGPT-4 appears feasible as an ePRO analysis tool for neuropathic pain and sleep problems in diabetes, warranting multicentre implementation studies evaluating clinical outcomes, patient experience, and workload effects under governance. Relevance to clinical practice: ChatGPT-4 can rapidly summarize diabetes ePROs on neuropathic pain and sleep to support triage, flag safety concerns, and streamline documentation, provided that clinician oversight, persona-standardized prompting, and appropriate governance are in place in routine care.]]>
Copyrights © 2026
